Terence Tao's Lecture: AI Excels at Problem-Solving but Lacks Evidence of Building New Theories

GaryMarcus · x · 2026-08-04

Gary Marcus highly recommends Terence Tao's recent lecture on AI and mathematics. Tao highlights two key risks: we are facing "proof indigestion" where AI generates a surplus of proofs that aren't all useful, and mathematics is not a uniform field—AI may excel at specific tasks but cannot cover the entire spectrum of mathematical research.

Echoing this view, Marcus evaluates Google's newly announced math AI model Astra, noting that while it demonstrates strong problem-solving capabilities, there is currently no evidence that such models can construct entirely new mathematical theories.

Related event: Terence Tao: AI Excels at Problem-Solving but Lacks Evidence for New Theories(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →